Analysis of the Effect of Object Scale on Grasp Generation Using GraspLDM

Figure 1.Types of text-guided data generation for embodied AI.Static scene generation often prevents task completion due to object positioning, regardless of the task requirements (left).Action generation is limited by a lack of interaction diversity, failing to account for the effects of surrounding environments (middle).To the best of our knowledge, we are the first to propose a method for dynamic scene generation that enables more realistic and diverse task interactions by the joint generation of static scenes and robotic actions (right).